When will these people get it? Telephotos CAN shrink

In rec.photo.digital.slr-systems David Dyer-Bennet wrote:
"Trevor" writes:

"Me" wrote in message
...
Anyway. if some sources are to be believed (Thom Hogan is one), a 36x24mm
sensor costs about $500 to produce. By the time margins are included,
that equates to about $1500 retail,

I simply cannot believe that, and would require some serious proof to
believe it is anything other than a *very small* fraction of that price if
produced in the quantities a cut price Canon or Nikon (not Sony) FF DSLR
would sell.

The full-frame sensor is *immensely* bigger than for example top-end
modern CPUs.

Sandy Bridge-EP-8 is 435 mm^2.

Andrew.

On 15/03/2012 14:28, Andrew Haley wrote:
In rec.photo.digital.slr-systems David wrote:
writes:

wrote in message
...
Anyway. if some sources are to be believed (Thom Hogan is one), a 36x24mm
sensor costs about $500 to produce. By the time margins are included,
that equates to about $1500 retail,

I simply cannot believe that, and would require some serious proof to
believe it is anything other than a *very small* fraction of that price if
produced in the quantities a cut price Canon or Nikon (not Sony) FF DSLR
would sell.

The full-frame sensor is *immensely* bigger than for example top-end
modern CPUs.

Sandy Bridge-EP-8 is 435 mm^2.

Andrew.

That is still only half the area of a full frame CCD sensor at 864mm^2.

--
Regards,
Martin Brown

On 3/15/2012 10:50 AM, Martin Brown wrote:
On 15/03/2012 14:28, Andrew Haley wrote:
In rec.photo.digital.slr-systems David wrote:
writes:

wrote in message
...
Anyway. if some sources are to be believed (Thom Hogan is one), a
36x24mm
sensor costs about $500 to produce. By the time margins are included,
that equates to about $1500 retail,

I simply cannot believe that, and would require some serious proof to
believe it is anything other than a *very small* fraction of that
price if
produced in the quantities a cut price Canon or Nikon (not Sony) FF
DSLR
would sell.

The full-frame sensor is *immensely* bigger than for example top-end
modern CPUs.

Sandy Bridge-EP-8 is 435 mm^2.

Andrew.

That is still only half the area of a full frame CCD sensor at 864mm^2.

True enough and I think it is worth adding that the Sandy Bridge CPU is
just about 21mm on a side as opposed to full frame at 36x24mm.

--
Jim Silverton

Extraneous "not" in Reply To.

On 2012-03-15 11:45 , James Silverton wrote:
On 3/15/2012 10:50 AM, Martin Brown wrote:

That is still only half the area of a full frame CCD sensor at 864mm^2.

True enough and I think it is worth adding that the Sandy Bridge CPU is
just about 21mm on a side as opposed to full frame at 36x24mm.

A photo sensor chip doesn't needs the fine layout of a CPU.

CPU's like the i7 are about 730M transistors per chip (quad core) on
263mm^2. That's nearly 3M per square mm. (!).

A FF photography sensor has up to 36M photo sites with up to 4
transistors per site (2 or 3 is typical). Edge logic will bump that up
a little but 2 orders of magnitude less. All in all, less than 150M
transistors on a chip that is over 3x the size of an i7.

A mere 200,000 transistors (at most) per mm^2.

Potentially lower costs per chip even considering defect rejection.

OTOH, the volume of CPU production is an order or 2 higher than FF
sensors leading to lower costs overall.

--
"I was gratified to be able to answer promptly, and I did.
I said I didn't know."
-Samuel Clemens.

On Thu, 15 Mar 2012 16:58:56 -0400, Alan Browne wrote:

On 2012-03-15 11:45 , James Silverton wrote:
On 3/15/2012 10:50 AM, Martin Brown wrote:

That is still only half the area of a full frame CCD sensor at
864mm^2.

True enough and I think it is worth adding that the Sandy Bridge CPU is
just about 21mm on a side as opposed to full frame at 36x24mm.

A photo sensor chip doesn't needs the fine layout of a CPU.

CPU's like the i7 are about 730M transistors per chip (quad core) on
263mm^2. That's nearly 3M per square mm. (!).

A FF photography sensor has up to 36M photo sites with up to 4
transistors per site (2 or 3 is typical). Edge logic will bump that up
a little but 2 orders of magnitude less. All in all, less than 150M
transistors on a chip that is over 3x the size of an i7.

A mere 200,000 transistors (at most) per mm^2.

Potentially lower costs per chip even considering defect rejection.

OTOH, the volume of CPU production is an order or 2 higher than FF
sensors leading to lower costs overall.

The density of circuits doesn't matter, other than the extent to which
they require newer fab lines. All that matters (regarding manufacturing
cost) is number of process steps and yield per wafer.

NRE costs on a processor are probably much higher than those on a sensor,
but the much higher volumes will help with that.

Processors have the large advantage (compared to sensors) that they now
have significant redundancy that allows the effective yield to be much
higher than the defect-free die rate: large areas are cache, and cache
blocks can be re-mapped around defects; multiple core parts can be sold
as reduced-core-count versions by marking defective cores inactive.

I expect that the price of 35mm sensors has been decreasing with time, as
one would expect, but they'd still be relatively expensive. Which is a
pitty: I'd love someone to make a 35mm rangefinder competitor to Leica
that I could afford.. I don't consider my D700 to have been outrageously
expensive, but I would definitely appreciate similar functionality in a
smaller body. Don't need an AF motor, and personally I don't need the
sophisticated metering and AF sensor electronics. Should be possible to
get a 35mm sensor into a body about as big as an FE2 or FM3a (or much
smaller, if you go for an EVF and do away with the mirror box and
pentaprism.)

Cheers,

--
Andrew

"Andrew Reilly" wrote:
On Thu, 15 Mar 2012 16:58:56 -0400, Alan Browne wrote:

A photo sensor chip doesn't needs the fine layout of a CPU.

CPU's like the i7 are about 730M transistors per chip (quad core) on
263mm^2. That's nearly 3M per square mm. (!).

FWIW, wiki claims 215 sq mm and 1.16 billion transistors.

A FF photography sensor has up to 36M photo sites with up to 4
transistors per site (2 or 3 is typical). Edge logic will bump that up
a little but 2 orders of magnitude less. All in all, less than 150M
transistors on a chip that is over 3x the size of an i7.

A mere 200,000 transistors (at most) per mm^2.

Potentially lower costs per chip even considering defect rejection.

OTOH, the volume of CPU production is an order or 2 higher than FF
sensors leading to lower costs overall.

The density of circuits doesn't matter, other than the extent to which
they require newer fab lines. All that matters (regarding manufacturing
cost) is number of process steps and yield per wafer.

Right. But those blokes are always busting their butts to improve yields.
The idea that there have been no significant improvements in SIX years is
beyond unbelievable.

Processors have the large advantage (compared to sensors) that they now
have significant redundancy that allows the effective yield to be much
higher than the defect-free die rate: large areas are cache, and cache
blocks can be re-mapped around defects; multiple core parts can be sold
as reduced-core-count versions by marking defective cores inactive.

Sure, but a fully functional quad-core i7 2600 is still only US$300. A
friggin billion functioning transistors all working at 3.4 GHz for a measly
US$300. Kewl beyond words.

But dSLR sensors are not defect free. They all have some number of dead
pixels, which are "mapped out" in demosaicing.

I expect that the price of 35mm sensors has been decreasing with time, as
one would expect, but they'd still be relatively expensive.

But I'd expect that many of the improvements in clean room and other
fabrication technologies that are absolutely necessary for the 64nm things
(that were coming online back when the 5D came out) to say nothing of the
current 32nm generation, would be applicable (and actually applied) in
bread-and-butter commercial foundries and non-bleeding-edge in-house lines.

The question is: how much does something that probably cost around US$500 to
produce in 2006 cost to produce in 2012? If it's down to 1/3, then it'd be
real easy to make a FF Rebel.

Which is a
pitty: I'd love someone to make a 35mm rangefinder competitor to Leica
that I could afford.. I don't consider my D700 to have been outrageously
expensive, but I would definitely appreciate similar functionality in a
smaller body. Don't need an AF motor, and personally I don't need the
sophisticated metering and AF sensor electronics. Should be possible to
get a 35mm sensor into a body about as big as an FE2 or FM3a (or much
smaller, if you go for an EVF and do away with the mirror box and
pentaprism.)

We're on the same page here. I've never used burst mode, my main lenses are
all manual focus, and AE can't possibly produce the right exposure all the
time, so for critical stuff you have to do something that does work (spot or
incident metering) or chimp.

--
David J. Littleboy
Tokyo, Japan

"Alan Browne" wrote in message
...
OTOH, the volume of CPU production is an order or 2 higher than FF sensors
leading to lower costs overall.

I would have thought far more than 2 times, but the number of FF sensors
made would increase rapidly if a cheaper FF camera becomes available.

Trevor.

In article , Trevor
wrote:

OTOH, the volume of CPU production is an order or 2 higher than FF sensors
leading to lower costs overall.

I would have thought far more than 2 times, but the number of FF sensors
made would increase rapidly if a cheaper FF camera becomes available.

right, because factories just magically appear as needed.

Andrew Reilly writes:

The density of circuits doesn't matter, other than the extent to which
they require newer fab lines. All that matters (regarding manufacturing
cost) is number of process steps and yield per wafer.

I don't think that's true. The smaller the detail size, the smaller a
wafer flaw becomes a device flaw, for example.
--
David Dyer-Bennet, ; http://dd-b.net/
Snapshots: http://dd-b.net/dd-b/SnapshotAlbum/data/
Photos: http://dd-b.net/photography/gallery/
Dragaera: http://dragaera.info

In article , says...

Andrew Reilly writes:

The density of circuits doesn't matter, other than the extent to which
they require newer fab lines. All that matters (regarding manufacturing
cost) is number of process steps and yield per wafer.

I don't think that's true. The smaller the detail size, the smaller a
wafer flaw becomes a device flaw, for example.

And that goes into yield per wafer. You expect a certain number of
rejects. If you're making one part per wafer that flaw means that you
reject an entire wafer's worth of production. If you're making 1000
parts per wafer that flaw may affect one or two or three or four of them
and you've lost 1/250 of a wafer's worth of production.